Phylogeny By Top Down Clustering Using a Given Multiple Alignment

We present a new phylogenetic tree construction algorithm. Our algorithm takes as input an alignment of multiple sequences. We assume that the sequences in the given alignment are closely related by evolution, and that the given alignment is biologically correct. Our approach has some similarities with the well-known clustering procedure UPGMA in that we define the distance between two clusters similarly, and we can build the tree using vertical distances to denote evolutionary distances. However, our approach is fundamentally different in choosing the clusters since it works from top to bottom. We select a pair of clusters with the maximum distance first, and then recursively create clusters for each. Since trying all possible clusterings is impractical, as an approximation to this approach, we examine all conserved segments in the given multiple sequence alignment, and for each we try two clusters: one in which the conserved region is identical in each sequence in this cluster, and the cluster composed of the rest of the sequences. Our main hypothesis is that the top-down approach will perform better compared to a bottom-up UPGMA method, in separating large cluster pairs from each other, and this will not prevent very close sequences from appearing in near leaves. We tested our algorithm using a multiple sequence alignment given in a recent paper that builds a phylogenetic tree of amino acid sequences from the Runt domain of RUNX genes from a set of species using existing techniques. Our tree has a few differences compared to this tree. We believe that our approach provides alternative valuable information for phylogeny.